loqrecovery: TODO

Epic: none
Release note: none

Author: pav-kv

pkg/kv/kvserver/loqrecovery/apply.go

@@ -317,6 +317,8 @@ func applyReplicaUpdate(
 
 	hs.LeadEpoch = 0
 
+	// TODO(sep-raft-log): when raft and state machine engines are separated, this
+	// update must be written to the raft engine.
 	if err := sl.SetHardState(ctx, readWriter, hs); err != nil {
 		return PrepareReplicaReport{}, errors.Wrap(err, "setting HardState")
 	}

pkg/kv/kvserver/loqrecovery/collect.go

@@ -192,6 +192,9 @@ func visitStoreReplicas(
 		if err != nil {
 			return err
 		}
+		// TODO(pav-kv): the LoQ recovery flow uses only the applied index, and the
+		// HardState.Commit loaded here is unused. Consider removing. Make sure this
+		// doesn't break compatibility for ReplicaInfo unmarshalling.
 		hstate, err := rsl.LoadHardState(ctx, raft)
 		if err != nil {
 			return err
@@ -201,6 +204,10 @@ func visitStoreReplicas(
 		// at potentially uncommitted entries as we have no way to determine their
 		// outcome, and they will become committed as soon as the replica is
 		// designated as a survivor.
+		// TODO(sep-raft-log): decide which LogID to read from. If the raft and
+		// state machine readers are slightly out of sync, the LogIDs may mismatch.
+		// For the heuristics here, it would probably make sense to read from the
+		// latest LogID in the raft engine.
 		rangeUpdates, err := GetDescriptorChangesFromRaftLog(
 			ctx, desc.RangeID, rstate.RaftAppliedIndex+1, math.MaxInt64, raft)
 		if err != nil {

pkg/kv/kvserver/loqrecovery/server.go

@@ -136,8 +136,9 @@ func (s Server) ServeLocalReplicas(
 		g.Go(func() error {
 			// TODO(sep-raft-log): when raft and state machine engines are separate,
 			// we need two snapshots here. This path is online, so we should make sure
-			// these snapshots are consistent. In particular, the LogID must match
-			// across the two.
+			// these snapshots are consistent (in particular, the LogID must match
+			// across the two), or make sure that LogID mismatch is handled in
+			// visitStoreReplicas.
 			reader := s.TODOEngine().NewSnapshot()
 			defer reader.Close()
 			return visitStoreReplicas(ctx, reader, reader, s.StoreID(), s.NodeID(),